Transcript Color spaces and JPEG

Color spaces and JPEG
Colors



physically, color is electro-magnetic radiation (i.e.
light with various wave length, between 390nm750nm) percieved by the human eye
a color is actually made from a combination o light
radiations with different wave lengths
electro-magnetic radiation spectrum:
Color spaces






color space = a mathematical model used to describe
colors as tuples of numbers
RGB – Red, Green, Blue
CMYK – Cyan, Magenta, Yellow, Key Black
YUV(YCbCr) – Luminance, Chrominance blue,
Chrominance red
HSV (HSB) – Hue, Saturation, Value
HSL – Hue, Saturation, Lightness
RGB






the color is specified as an additive combination of
three primary colors: Red, Green, Blue
in addition, a white point must be specified for this
color model
is mostly used in computer graphics
has a variation, RGBA, with alpha channel for
transparency
black is 0 0 0
white is 255 255 255
CMYK


the color is obtained by substracting from a white
substrate the color components cyan, magenta, yellow
and black
mainly used in paper printing
YUV


a color is composed from 3 components: Y-luminance
(brightness of the pixel), U-blue chrominance, V-red
chrominance
YPbPr is a scaled version of YUV used in analog
television standards and YCbCr is a scaled version of
YUV used in digital films and video and image
compression standards like MPEG and JPEG
HSL and HSV




a color is described by 3 components: Hue (nuanta de
culoare), Saturation(saturatia culorii) and
lightness/brightness
HSV is also known as HSB (hue, saturation, brightness)
is mainly used by artists
HSL and HSV are cylindrical-coordinate representation
of color points in the RGB (cartezian-coordinate) model
Color space conversions

RGB to YCbCr
Y = 0.299*R + 0.587*G + 0.114*B
Cb = 128 – 0.1687*R – 0.3312*G + 0.5*B
Cr = 128 + 0.5*R – 0.4186*G – 0.0813*B

RGB (1-255) to CMY (0-1)
C = 1 – (R / 255)
M = 1 – (G / 255)
Y = 1 – (B / 255)
JPEG - Joint Photographic
Experts Group
JPEG




is the name of an digital image compression standard
created by ISO/IEC Joint Technical Committee 1,
Subcommittee 29, Working Group 1 (ISO/IEC JTC 1/SC
29/WG 1); the standard is also recommended by ITU-T
stands from Joint Photographic Experts Group, the
name of the committee that created the standard
is a lossy compression standard (different than lossless
image compression like TIFF, GIF, PNG, BMP etc.)
JPEG has 2 operation modes:


baseline – lossy compression with a quality/compression factor
from 1 to 100
progressive – an image is compressed in multiple phases of
progressively higher detail
JPEG baseline process


JPEG operates on 8x8 or 16x16 pixels macroblocks
which are compressed independently
the JPEG encoder/decoder structure:
JPEG baseline compression algorithm
1.
2.
3.
4.
5.
Color space conversion (to YUV) and possibly padding
Downsampling & block splitting
Discrete Cosine Transform (DCT)
Quantization
Entropy encoding
5.1 Zig-zag order of the coefficients, then Run-length
encoding
5.2 Huffman encoding
1. Color space conversion (to YUV)
and possibly padding


the colors of pixels are converted to YUV color space
Y = 0.299*R + 0.587*G + 0.114*B
U = 128 – 0.1687*R – 0.3312*G + 0.5*B
V = 128 + 0.5*R – 0.4186*G – 0.0813*B
then the image is pixel padded at right and bottom
so that width and height are multiple of 8 (16) bits
2. Downsampling & block splitting

YUV image is split in 8x8 or 16x16 blocks and
downsampled:



4:4:4
4:2:2
4:0:0
3. Discrete Cosine Transform (DCT)

the color values (YUV) are converted from the spatial
(time) domain into frequency domain using the DCT
formula bellow (similar to DFT – Discrete Fourier
Transform):
7
7
1
 (2 x  1)u 
 (2 y  1)v 
F (u, v)  c(u)c(v) f ( x, y) * cos
 * cos

4
16
16




x 0 y 0
f(x,y) – pixel color (x=0..7, y=0..7)
c(u)=c(v)=1/sqrt(2) for u,v=0
c(u)=c(v)=1 otherwise
F(0,0) - DC coefficient
F(u,v) – AC coefficients (u,v different than 0)
3. Discrete Cosine Transform (2)
3. Discrete Cosine Transform (3)



Each 8x8 block of source image samples is effectively a 64point discrete signal which is a function of the two spatial
dimensions x and y. The DCT takes such a signal as its
input and decomposes it into 64 orthogonal basis signals.
Each contains one of the 64 unique two-dimensional (2D)
“spatial frequencies’’ which comprise the input signal’s
“spectrum.” The output of the DCT is the set of 64 basissignal amplitudes or “DCT coefficients” whose values are
uniquely determined by the particular 64-point input signal
the DCT tends to concentrate the strength (i.e. average
intensity/color) of the block in the DC coefficient (the coef.
of zero frequency in both dimensions;
the other coefficients contain variations of the average
intensity/color and are called AC coefficients
4. Quantization

each DCT coefficient obtained at step 3 is divided by
a quantization value
5. Entropy encoding
 entropy enoding = zig-zag order + run-length encoding
+ Huffman encoding
 Zig-zag order:
The preceeding block is encoded as: 150, 80, 92, 26, 75, 20, 4, 18, 19, 3,
1, 2, 13, 3, 1, 0, 1, 2, 2, 0, 0, 0, 0, 0, 1, 1, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0,
0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0
Entropy encoding of the DC coef.




the DC coefficient of a block is encoded separately
than the AC coefficients of that block
the difference between the current DC and the DC
from the previous block is encoded as 2 symbols:
(SIZE) (AMPLITUDE)
SIZE = is the number of bits used to encode
AMPLITUDE; is encoded as a variable-length
code(VLC) from a Huffman table
AMPLITUDE = is the amplitude on the coefficient
difference; is encoded as a variable-length integer
(VLI) code whose length in bits is given in the table
from the next slide
The table for the VLI code of AMPLITUDE
Entropy encoding of the AC coefs.


AC coefficients are parsed in a zig-zag order and then run-length
encoded and then Huffman encoded
in general, the sequence of characters:
a b c c c c d d d e f g g g g g g h h is run-length encoded into the sequence:
a b 4c 3d e f 6g 2h



in JPEG, each non-zero AC coef. is encoded in combination with the
runlength (consecutive number) of zero-valued AC coefs. into a pair
of symbols:
(RUNLENGTH, SIZE) (AMPLITUDE)
where SIZE and AMPLITUDE are like the ones used for the DC coef.
and RUNLENGTH – the number of consecutive zero-valued AC
coefs. in zig-zag order preceeding the nonzero AC coef. being
represented
symbol 1 is encoded as a variable-length code(VLC) from a
Huffman table
symbol 2 is encoded as a variable-length integer (VLI) code whose
length in bits is given in the previous table
Entropy encoding of the previous
quantization block example

the zig-zag order of coefficients:
150, 80, 92, 26, 75, 20, 4, 18, 19, 3, 1, 2, 13, 3, 1, 0, 1, 2, 2, 0, 0, 0, 0, 0, 1, 1, 0, 0,
0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0,
0, 0, 0
is run-length encoded (assume DC in the previous block is
0) into:
(8)(150), (0,7)(80), (0,7)(92), (0,5)(26), (0,7)(75), (0,5)(20), (0,3)(4), (0,5)(18),
(0,5)(19), (0,2)(3), (0,1)(1), (0,2)(2), (0,4)(13), (0,2)(3), (0,1)(1), (1,1)(1), (0,2)(2),
(0,2)(2), (5,1)(1), (0,1)(1), (0,0)

(0,0) – is EOB (End Of Block)
the above sequence is Huffman (VLC and VLI) encoded
into: (111110)(10010110), (11111000)(1010000), (11111000)(1011100),
(11010)(11010), (11111000)(1001011), (11010)(10100), (100)(100), (11010)(10010),
(11010)(10011), (01)(11), (00)(1), (01)(10), (1011)(1101), (01)(11), (00)(1),
(1100)(1), (01)(10), (01)(10), (1111010)(1), (00)(1), (1010)
JFIF File Format (.jpg) –
see class
dmms.jpeg.JPGInfo.java
The format of a JPEG/JFIF file is:

Header:



Segments or markers:




It occupies two bytes.
0xff, 0xd8 (SOI : Start Of Image ) (these two identify a JPEG/JFIF file).
Following the SOI marker, there can be any number of segments or
markers such as:
APP0..APP15, SOF0..SOF15, DQT, DHT, SOS, JPG, JPG0..JPG13, DAC,
DNL, DRI, DHP, EXP, RST0..RST7, TEM, COM.
An APP0 segment immediately follows the SOI marker.
Trailer:


It occupies two bytes.
0xff, 0xd9 (EOI: End of Image) (these two identify end of image).
Note: any number of 0xff bytes between two segments (markers) must be ignored.